A Low Complexity Extended Kalman Filter Algorithm for Neural Network Digital Predistortion of Power Amplifier

Abstract

In advanced wireless communication systems that require spectrally efficient modulation schemes, the modulated signal with a high peak-to-average power ratio (PAPR) drives the power amplifier (PA) to operate near the saturation region and introduces serious nonlinearity of the PA. Digital predistortion (DPD) is one of the most promising techniques for PA linearization. In this paper, we propose a low complexity extended Kalman filter (LC-EKF) algorithm for training a neural network (NN) in the design of a predistorter for a DPD system. We propose a method to decrease the matrix dimensions during the matrix inversion computation of the Kalman gain using the matrix inversion lemma, thereby reducing the complexity of the implementation of the EKF algorithm. We evaluate the proposed LC-EKF algorithm for the neural network DPD system in terms of both complexity and performance. The simulation results show that the proposed LCEKF algorithm has considerably less complexity than the traditional EKF algorithm without sacrificing its performance and better normalized mean squared error (NMSE) and adjacent channel power ratio (ACPR) performance than the Levenberg- Marquardt (LM) trained neural network in the DPD system.